Detection of maize seed viability using time series multispectral imaging technology

Abstract

Maize seed vigor significantly impacts seedling emergence and overall yield. Thus, accurately assessing seed viability is crucial for ensuring crop quality. This study employs multispectral imaging to capture spectral images of maize seeds during the swelling absorption process. We analyzed and compared the spectral characteristics and their trends between viability and non-viability seeds across various absorption times, specifically at 12-h intervals. To improve the identification of seed viability, we integrated spectral data collected at multiple water absorption times with spectral difference data at 12-h intervals, forming a comprehensive time-series dataset. A classification model for seed viability was developed using stochastic subspace screening in conjunction with support vector machine (SVM) techniques. The results indicate that the stochastic subspace integrated learning approach effectively classifies maize seed viability, achieving classification accuracy exceeding 90 % after 36 h of water absorption. This method enables viability detection prior to seed germination. In conclusion, the integration of stochastic subspace learning and time-series spectral data significantly improves the identification of maize seed viability, offering new insights for seed viability detection.

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.